Google Launches ARD Open Spec for AI Agent Tool Discovery

When an operations agent investigates a production incident, it might need to query observability systems, search documentation, review deployment history, and consult specialized agents, each registered in a different place by a different team. According to Help Net Security, Google's new Agentic Resource Discovery (ARD) specification is designed to replace that fragmentation with a standard protocol where agents can find and verify capabilities across organizational boundaries at runtime.

The problem the spec frames is precise. As Google's software engineers put it, "agents need reliable answers to three questions: Where does the right capability live? Which capability should I actually use? And how do I verify it's safe to connect to?" ARD answers all three with two primitives. Organizations publish an ai-catalog.json file at a well-known domain path listing MCP servers, A2A agents, OpenAPI tools, or nested catalogs, with publisher metadata enabling cryptographic identity verification. Registries then function as search engines for the agentic web, crawling those catalogs and returning ranked, verified matches when agents submit discovery requests.

The specification is released under the Apache 2.0 license and builds on the AI Catalog data model from the AI Catalog Working Group under the Linux Foundation. Google Cloud plans to support ARD through Agent Registry in the Gemini Enterprise Agent Platform, adding enterprise features like globally unique namespaced URNs, agentic egress policy enforcement, and HIPAA compliance, though the reporting notes native ARD support for Agent Platform will arrive in the coming months rather than being available today.

The familiar tension in "open spec plus proprietary hosted registry" applies here: the spec stays open, but if most enterprises take Google Cloud's Agent Registry as the path of least resistance, the terms of that managed service effectively set the rules. What the reporting also does not give you is any named enterprise partners who have committed to publishing ARD catalogs, which leaves the network-effects story unproven. Discovery registries are only as useful as the catalogs they index.

For tool authors and platform engineers building agents that cross team or organizational lines, the upside is straightforward: discoverability without custom integration work for each consumer, and a structured audit trail that compliance teams can actually use. A quickstart is available at agenticresourcediscovery.org. The practical question over the next year is less about the protocol's technical soundness and more about whether enough organizations publish catalogs to make registry queries worth running.